Skip to contentSkip to conversation

acpi/gpex: Inform os to keep firmware resource map

PATCH In review Jiahui Cen 5 participants 11 messages Dec 17, 2020 – Jul 22, 2021

lore
Share & tools
lore

Jiahui Cen

author
#0 lore raw

PATCH 1/1 acpi/gpex: Inform os to keep firmware resource map

There may be some differences in pci resource assignment between guest os and firmware.

Eg. A Bridge with Bus [d2]

    -+-[0000:d2]---01.0-[d3]----01.0

    where [d2:01.00] is a pcie-pci-bridge with BAR0 (mem, 64-bit, non-pref) [size=256]
          [d3:01.00] is a PCI Device with BAR0 (mem, 64-bit, pref) [size=128K]
                                          BAR4 (mem, 64-bit, pref) [size=64M]

    In EDK2, the Resource Map would be:
        PciBus: Resource Map for Bridge [D2|01|00]
        Type = PMem64; Base = 0x8004000000;     Length = 0x4100000;     Alignment = 0x3FFFFFF
           Base = 0x8004000000; Length = 0x4000000;     Alignment = 0x3FFFFFF;  Owner = PCI [D3|01|00:20]
           Base = 0x8008000000; Length = 0x20000;       Alignment = 0x1FFFF;    Owner = PCI [D3|01|00:10]
        Type =  Mem64; Base = 0x8008100000;     Length = 0x100; Alignment = 0xFFF

    While in Linux, kernel will use 0x2FFFFFF as the alignment to calculate
    the PMem64 size, which would be 0x6000000.

The diffences could result in resource assignment failure.

Using _DSM #5 method to inform guest os not to ignore the PCI configuration that firmware has done at boot time could handle the differences.

---
 hw/pci-host/gpex-acpi.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)
hw/pci-host/gpex-acpi.c +10 -1
index 071aa11b5c..2b490f3379 100644
@@ -112,10 +112,19 @@ static void acpi_dsdt_add_pci_osc(Aml *dev)
     UUID = aml_touuid("E5C937D0-3553-4D7A-9117-EA4D19C3434D");
     ifctx = aml_if(aml_equal(aml_arg(0), UUID));
     ifctx1 = aml_if(aml_equal(aml_arg(2), aml_int(0)));
-    uint8_t byte_list[1] = {1};
+    uint8_t byte_list[1] = {0x21};
     buf = aml_buffer(1, byte_list);
     aml_append(ifctx1, aml_return(buf));
     aml_append(ifctx, ifctx1);
+
+    /* PCI Firmware Specification 3.2
+     * 4.6.5. _DSM for Ignoring PCI Boot Configurations
+     * The UUID in _DSM in this context is
+     * {E5C937D0-3553-4D7A-9117-EA4D19C3434D}
+     */
+    ifctx1 = aml_if(aml_equal(aml_arg(2), aml_int(5)));
+    aml_append(ifctx1, aml_return(aml_int(0)));
+    aml_append(ifctx, ifctx1);
     aml_append(method, ifctx);
 
     byte_list[0] = 0;
Signature · 1 line
2.28.0
author
#1 lore raw

+Laszlo

Quoted 56 lines · Jiahui Cen

There may be some differences in pci resource assignment between guest os and firmware.

Eg. A Bridge with Bus [d2]

    -+-[0000:d2]---01.0-[d3]----01.0

    where [d2:01.00] is a pcie-pci-bridge with BAR0 (mem, 64-bit, non-pref) [size=256]
          [d3:01.00] is a PCI Device with BAR0 (mem, 64-bit, pref) [size=128K]
                                          BAR4 (mem, 64-bit, pref) [size=64M]

    In EDK2, the Resource Map would be:
        PciBus: Resource Map for Bridge [D2|01|00]
        Type = PMem64; Base = 0x8004000000;     Length = 0x4100000;     Alignment = 0x3FFFFFF
           Base = 0x8004000000; Length = 0x4000000;     Alignment = 0x3FFFFFF;  Owner = PCI [D3|01|00:20]
           Base = 0x8008000000; Length = 0x20000;       Alignment = 0x1FFFF;    Owner = PCI [D3|01|00:10]
        Type =  Mem64; Base = 0x8008100000;     Length = 0x100; Alignment = 0xFFF

    While in Linux, kernel will use 0x2FFFFFF as the alignment to calculate
    the PMem64 size, which would be 0x6000000.

The diffences could result in resource assignment failure.

Using _DSM #5 method to inform guest os not to ignore the PCI configuration that firmware has done at boot time could handle the differences.

---
 hw/pci-host/gpex-acpi.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)
hw/pci-host/gpex-acpi.c +10 -1
index 071aa11b5c..2b490f3379 100644
@@ -112,10 +112,19 @@ static void acpi_dsdt_add_pci_osc(Aml *dev)
     UUID = aml_touuid("E5C937D0-3553-4D7A-9117-EA4D19C3434D");
     ifctx = aml_if(aml_equal(aml_arg(0), UUID));
     ifctx1 = aml_if(aml_equal(aml_arg(2), aml_int(0)));
-    uint8_t byte_list[1] = {1};
+    uint8_t byte_list[1] = {0x21};
     buf = aml_buffer(1, byte_list);
     aml_append(ifctx1, aml_return(buf));
     aml_append(ifctx, ifctx1);
+
+    /* PCI Firmware Specification 3.2
+     * 4.6.5. _DSM for Ignoring PCI Boot Configurations
+     * The UUID in _DSM in this context is
+     * {E5C937D0-3553-4D7A-9117-EA4D19C3434D}
+     */
+    ifctx1 = aml_if(aml_equal(aml_arg(2), aml_int(5)));
+    aml_append(ifctx1, aml_return(aml_int(0)));
+    aml_append(ifctx, ifctx1);
     aml_append(method, ifctx);
 
     byte_list[0] = 0;
#2 lore raw
Quoted 57 lines · Jiahui Cen

+Laszlo

Quoted 54 lines · Jiahui Cen

There may be some differences in pci resource assignment between guest os and firmware.

Eg. A Bridge with Bus [d2]

    -+-[0000:d2]---01.0-[d3]----01.0

    where [d2:01.00] is a pcie-pci-bridge with BAR0 (mem, 64-bit, non-pref) [size=256]
          [d3:01.00] is a PCI Device with BAR0 (mem, 64-bit, pref) [size=128K]
                                          BAR4 (mem, 64-bit, pref) [size=64M]

    In EDK2, the Resource Map would be:
        PciBus: Resource Map for Bridge [D2|01|00]
        Type = PMem64; Base = 0x8004000000;     Length = 0x4100000;     Alignment = 0x3FFFFFF
           Base = 0x8004000000; Length = 0x4000000;     Alignment = 0x3FFFFFF;  Owner = PCI [D3|01|00:20]
           Base = 0x8008000000; Length = 0x20000;       Alignment = 0x1FFFF;    Owner = PCI [D3|01|00:10]
        Type =  Mem64; Base = 0x8008100000;     Length = 0x100; Alignment = 0xFFF

    While in Linux, kernel will use 0x2FFFFFF as the alignment to calculate
    the PMem64 size, which would be 0x6000000.

The diffences could result in resource assignment failure.

Using _DSM #5 method to inform guest os not to ignore the PCI configuration that firmware has done at boot time could handle the differences.

---
 hw/pci-host/gpex-acpi.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)
hw/pci-host/gpex-acpi.c +10 -1
index 071aa11b5c..2b490f3379 100644
@@ -112,10 +112,19 @@ static void acpi_dsdt_add_pci_osc(Aml *dev)
     UUID = aml_touuid("E5C937D0-3553-4D7A-9117-EA4D19C3434D");
     ifctx = aml_if(aml_equal(aml_arg(0), UUID));
     ifctx1 = aml_if(aml_equal(aml_arg(2), aml_int(0)));
-    uint8_t byte_list[1] = {1};
+    uint8_t byte_list[1] = {0x21};
     buf = aml_buffer(1, byte_list);
     aml_append(ifctx1, aml_return(buf));
     aml_append(ifctx, ifctx1);
+
+    /* PCI Firmware Specification 3.2
+     * 4.6.5. _DSM for Ignoring PCI Boot Configurations
+     * The UUID in _DSM in this context is
+     * {E5C937D0-3553-4D7A-9117-EA4D19C3434D}
+     */
+    ifctx1 = aml_if(aml_equal(aml_arg(2), aml_int(5)));
+    aml_append(ifctx1, aml_return(aml_int(0)));
+    aml_append(ifctx, ifctx1);
     aml_append(method, ifctx);
 
Quoted 3 lines

hw/pci-host/gpex-acpi.c static void acpi_dsdt_add_pci_osc(Aml *dev) · line 130

Quoted 2 lines
hw/pci-host/gpex-acpi.c +0 -0
     byte_list[0] = 0;

Seems to make sense to me (I didn't realize we already had the _DSM method with this GUID!), but now I'm not sure what to expect of the guest kernel, in light of what Ard said. So if it works now, is that by accident, or is it an intentional, fresh commit in the kernel? Like a78cf9657ba5 ("PCI/ACPI: Evaluate PCI Boot Configuration _DSM", 2019-06-21)?

Benjamin: can you please tell us something about this Linux commit? What was the motivation for it?

Hmm.... this commit seems to be a part of the following series:

a78cf9657ba5 PCI/ACPI: Evaluate PCI Boot Configuration _DSM 7ac0d094fbe9 PCI: Don't auto-realloc if we're preserving firmware config 3e8ba9686600 arm64: PCI: Allow resource reallocation if necessary 85dc04136e86 arm64: PCI: Preserve firmware configuration when desired

OK, after reading through the commit messages in those commits (esp. 7ac0d094fbe9), I think the Linux change was made exactly for the purpose that we want it for -- stick with the firmware assignments.

Ard, does that seem right, or am I misunderstanding the kernel series?

Thanks
Laszlo

#3 lore raw
Quoted 78 lines · Laszlo Ersek
Quoted 57 lines · Jiahui Cen

+Laszlo

Quoted 54 lines · Jiahui Cen

There may be some differences in pci resource assignment between guest os and firmware.

Eg. A Bridge with Bus [d2]

    -+-[0000:d2]---01.0-[d3]----01.0

    where [d2:01.00] is a pcie-pci-bridge with BAR0 (mem, 64-bit, non-pref) [size=256]
          [d3:01.00] is a PCI Device with BAR0 (mem, 64-bit, pref) [size=128K]
                                          BAR4 (mem, 64-bit, pref) [size=64M]

    In EDK2, the Resource Map would be:
        PciBus: Resource Map for Bridge [D2|01|00]
        Type = PMem64; Base = 0x8004000000;     Length = 0x4100000;     Alignment = 0x3FFFFFF
           Base = 0x8004000000; Length = 0x4000000;     Alignment = 0x3FFFFFF;  Owner = PCI [D3|01|00:20]
           Base = 0x8008000000; Length = 0x20000;       Alignment = 0x1FFFF;    Owner = PCI [D3|01|00:10]
        Type =  Mem64; Base = 0x8008100000;     Length = 0x100; Alignment = 0xFFF

    While in Linux, kernel will use 0x2FFFFFF as the alignment to calculate
    the PMem64 size, which would be 0x6000000.

The diffences could result in resource assignment failure.

Using _DSM #5 method to inform guest os not to ignore the PCI configuration that firmware has done at boot time could handle the differences.

---
 hw/pci-host/gpex-acpi.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)
hw/pci-host/gpex-acpi.c +10 -1
index 071aa11b5c..2b490f3379 100644
@@ -112,10 +112,19 @@ static void acpi_dsdt_add_pci_osc(Aml *dev)
     UUID = aml_touuid("E5C937D0-3553-4D7A-9117-EA4D19C3434D");
     ifctx = aml_if(aml_equal(aml_arg(0), UUID));
     ifctx1 = aml_if(aml_equal(aml_arg(2), aml_int(0)));
-    uint8_t byte_list[1] = {1};
+    uint8_t byte_list[1] = {0x21};
     buf = aml_buffer(1, byte_list);
     aml_append(ifctx1, aml_return(buf));
     aml_append(ifctx, ifctx1);
+
+    /* PCI Firmware Specification 3.2
+     * 4.6.5. _DSM for Ignoring PCI Boot Configurations
+     * The UUID in _DSM in this context is
+     * {E5C937D0-3553-4D7A-9117-EA4D19C3434D}
+     */
+    ifctx1 = aml_if(aml_equal(aml_arg(2), aml_int(5)));
+    aml_append(ifctx1, aml_return(aml_int(0)));
+    aml_append(ifctx, ifctx1);
     aml_append(method, ifctx);
 
Quoted 3 lines

hw/pci-host/gpex-acpi.c static void acpi_dsdt_add_pci_osc(Aml *dev) · line 130

Quoted 2 lines
hw/pci-host/gpex-acpi.c +0 -0
     byte_list[0] = 0;

Seems to make sense to me (I didn't realize we already had the _DSM method with this GUID!), but now I'm not sure what to expect of the guest kernel, in light of what Ard said. So if it works now, is that by accident, or is it an intentional, fresh commit in the kernel? Like a78cf9657ba5 ("PCI/ACPI: Evaluate PCI Boot Configuration _DSM", 2019-06-21)?

Benjamin: can you please tell us something about this Linux commit? What was the motivation for it?

Hmm.... this commit seems to be a part of the following series:

a78cf9657ba5 PCI/ACPI: Evaluate PCI Boot Configuration _DSM 7ac0d094fbe9 PCI: Don't auto-realloc if we're preserving firmware config 3e8ba9686600 arm64: PCI: Allow resource reallocation if necessary 85dc04136e86 arm64: PCI: Preserve firmware configuration when desired

Quoted 6 lines

OK, after reading through the commit messages in those commits (esp. 7ac0d094fbe9), I think the Linux change was made exactly for the purpose that we want it for -- stick with the firmware assignments.

Ard, does that seem right, or am I misunderstanding the kernel series?

Hmm, I had no recollection of those changes going in, but I was clearly aware at the time, given my acks.

So we clearly do support DSM #5 to prevent resource reallocation, and it makes sense to use it in the proposed way.

#4 lore raw
In reply to Jiahui Cen
Quoted 20 lines · Jiahui Cen

There may be some differences in pci resource assignment between guest os and firmware.

Eg. A Bridge with Bus [d2]

    -+-[0000:d2]---01.0-[d3]----01.0

    where [d2:01.00] is a pcie-pci-bridge with BAR0 (mem, 64-bit, non-pref) [size=256]
          [d3:01.00] is a PCI Device with BAR0 (mem, 64-bit, pref) [size=128K]
                                          BAR4 (mem, 64-bit, pref) [size=64M]

    In EDK2, the Resource Map would be:
        PciBus: Resource Map for Bridge [D2|01|00]
        Type = PMem64; Base = 0x8004000000;     Length = 0x4100000;     Alignment = 0x3FFFFFF
           Base = 0x8004000000; Length = 0x4000000;     Alignment = 0x3FFFFFF;  Owner = PCI [D3|01|00:20]
           Base = 0x8008000000; Length = 0x20000;       Alignment = 0x1FFFF;    Owner = PCI [D3|01|00:10]
        Type =  Mem64; Base = 0x8008100000;     Length = 0x100; Alignment = 0xFFF

    While in Linux, kernel will use 0x2FFFFFF as the alignment to calculate
    the PMem64 size, which would be 0x6000000.
Quoted 1 line

The diffences could result in resource assignment failure.

A bit more data here please. When does this result in a failure?

Quoted 35 lines

Using _DSM #5 method to inform guest os not to ignore the PCI configuration that firmware has done at boot time could handle the differences.

---
 hw/pci-host/gpex-acpi.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)
hw/pci-host/gpex-acpi.c +10 -1
index 071aa11b5c..2b490f3379 100644
@@ -112,10 +112,19 @@ static void acpi_dsdt_add_pci_osc(Aml *dev)
     UUID = aml_touuid("E5C937D0-3553-4D7A-9117-EA4D19C3434D");
     ifctx = aml_if(aml_equal(aml_arg(0), UUID));
     ifctx1 = aml_if(aml_equal(aml_arg(2), aml_int(0)));
-    uint8_t byte_list[1] = {1};
+    uint8_t byte_list[1] = {0x21};
     buf = aml_buffer(1, byte_list);
     aml_append(ifctx1, aml_return(buf));
     aml_append(ifctx, ifctx1);
+
+    /* PCI Firmware Specification 3.2
+     * 4.6.5. _DSM for Ignoring PCI Boot Configurations
+     * The UUID in _DSM in this context is
+     * {E5C937D0-3553-4D7A-9117-EA4D19C3434D}
+     */
+    ifctx1 = aml_if(aml_equal(aml_arg(2), aml_int(5)));
+    aml_append(ifctx1, aml_return(aml_int(0)));
+    aml_append(ifctx, ifctx1);
     aml_append(method, ifctx);
 
     byte_list[0] = 0;
Signature · 1 line
2.28.0
author
#5 lore raw

Hi Michael,

Quoted 18 lines · Michael S. Tsirkin
Quoted 17 lines · Jiahui Cen

There may be some differences in pci resource assignment between guest os and firmware.

Eg. A Bridge with Bus [d2]

    -+-[0000:d2]---01.0-[d3]----01.0

    where [d2:01.00] is a pcie-pci-bridge with BAR0 (mem, 64-bit, non-pref) [size=256]
          [d3:01.00] is a PCI Device with BAR0 (mem, 64-bit, pref) [size=128K]
                                          BAR4 (mem, 64-bit, pref) [size=64M]

    In EDK2, the Resource Map would be:
        PciBus: Resource Map for Bridge [D2|01|00]
        Type = PMem64; Base = 0x8004000000;     Length = 0x4100000;     Alignment = 0x3FFFFFF
           Base = 0x8004000000; Length = 0x4000000;     Alignment = 0x3FFFFFF;  Owner = PCI [D3|01|00:20]
           Base = 0x8008000000; Length = 0x20000;       Alignment = 0x1FFFF;    Owner = PCI [D3|01|00:10]
        Type =  Mem64; Base = 0x8008100000;     Length = 0x100; Alignment = 0xFFF
Quoted 2 lines
Quoted 2 lines
    While in Linux, kernel will use 0x2FFFFFF as the alignment to calculate
    the PMem64 size, which would be 0x6000000.

Sorry, I made a mistake. Kernel will use 0x1FFFFFF (the half of 0x3FFFFFF) not 0x2FFFFFF as alignment, so the size 0x4100000 would be aligned to 0x6000000.

Quoted 5 lines
Quoted 2 lines

The diffences could result in resource assignment failure.

A bit more data here please. When does this result in a failure?

EDK2 would use 0x4100000 to calculate the root bus's PMem64 resource window, while Kernel would try to allocate 0x6000000 from the PMem64 resource window. Since the resource window is not large enough, the allocation would fail.

Is this clear enough? I would add this into my next patch.

Thanks,
Jiahui

Quoted 38 lines
Quoted 35 lines

Using _DSM #5 method to inform guest os not to ignore the PCI configuration that firmware has done at boot time could handle the differences.

---
 hw/pci-host/gpex-acpi.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)
hw/pci-host/gpex-acpi.c +10 -1
index 071aa11b5c..2b490f3379 100644
@@ -112,10 +112,19 @@ static void acpi_dsdt_add_pci_osc(Aml *dev)
     UUID = aml_touuid("E5C937D0-3553-4D7A-9117-EA4D19C3434D");
     ifctx = aml_if(aml_equal(aml_arg(0), UUID));
     ifctx1 = aml_if(aml_equal(aml_arg(2), aml_int(0)));
-    uint8_t byte_list[1] = {1};
+    uint8_t byte_list[1] = {0x21};
     buf = aml_buffer(1, byte_list);
     aml_append(ifctx1, aml_return(buf));
     aml_append(ifctx, ifctx1);
+
+    /* PCI Firmware Specification 3.2
+     * 4.6.5. _DSM for Ignoring PCI Boot Configurations
+     * The UUID in _DSM in this context is
+     * {E5C937D0-3553-4D7A-9117-EA4D19C3434D}
+     */
+    ifctx1 = aml_if(aml_equal(aml_arg(2), aml_int(5)));
+    aml_append(ifctx1, aml_return(aml_int(0)));
+    aml_append(ifctx, ifctx1);
     aml_append(method, ifctx);
 
     byte_list[0] = 0;
Signature · 1 line
2.28.0

.

#6 lore raw
In reply to Jiahui Cen
Quoted 35 lines · Jiahui Cen

There may be some differences in pci resource assignment between guest os and firmware.

Eg. A Bridge with Bus [d2]

    -+-[0000:d2]---01.0-[d3]----01.0

    where [d2:01.00] is a pcie-pci-bridge with BAR0 (mem, 64-bit, non-pref) [size=256]
          [d3:01.00] is a PCI Device with BAR0 (mem, 64-bit, pref) [size=128K]
                                          BAR4 (mem, 64-bit, pref) [size=64M]

    In EDK2, the Resource Map would be:
        PciBus: Resource Map for Bridge [D2|01|00]
        Type = PMem64; Base = 0x8004000000;     Length = 0x4100000;     Alignment = 0x3FFFFFF
           Base = 0x8004000000; Length = 0x4000000;     Alignment = 0x3FFFFFF;  Owner = PCI [D3|01|00:20]
           Base = 0x8008000000; Length = 0x20000;       Alignment = 0x1FFFF;    Owner = PCI [D3|01|00:10]
        Type =  Mem64; Base = 0x8008100000;     Length = 0x100; Alignment = 0xFFF

    While in Linux, kernel will use 0x2FFFFFF as the alignment to calculate
    the PMem64 size, which would be 0x6000000.

The diffences could result in resource assignment failure.

Using _DSM #5 method to inform guest os not to ignore the PCI configuration that firmware has done at boot time could handle the differences.

---
 hw/pci-host/gpex-acpi.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)
hw/pci-host/gpex-acpi.c +0 -0
index 071aa11b5c..2b490f3379 100644

hw/pci-host/gpex-acpi.c static void acpi_dsdt_add_pci_osc(Aml *dev) · line 116

Quoted 6 lines
hw/pci-host/gpex-acpi.c +1 -1
@@ -112,10 +112,19 @@ static void acpi_dsdt_add_pci_osc(Aml *dev)
     UUID = aml_touuid("E5C937D0-3553-4D7A-9117-EA4D19C3434D");
     ifctx = aml_if(aml_equal(aml_arg(0), UUID));
     ifctx1 = aml_if(aml_equal(aml_arg(2), aml_int(0)));
-    uint8_t byte_list[1] = {1};
+    uint8_t byte_list[1] = {0x21};
     buf = aml_buffer(1, byte_list);

Hmm what is this change for?

I also noticed something weird.
Spec seems to say for _DSM for PCI Express Slot Information:

Arguments:
Arg0: UUID: E5C937D0-3553-4d7a-9117-EA4D19C3434D
Arg1: Revision ID: 2
Arg2: Function Index: 1
Arg3: Empty Package

how come we are comparing function index to 0 here?

Also, as long as we are changing this probably shouldn't hard-code 1 as array size ...

hw/pci-host/gpex-acpi.c static void acpi_dsdt_add_pci_osc(Aml *dev) · line 121

Quoted 5 lines
hw/pci-host/gpex-acpi.c +3 -0
     aml_append(ifctx1, aml_return(buf));
     aml_append(ifctx, ifctx1);
+
+    /* PCI Firmware Specification 3.2
+     * 4.6.5. _DSM for Ignoring PCI Boot Configurations

Note you must always quote the most recent spec that your change refers to. This is so people can figure out legacy guest compatibility.

In this case I think this first appeard in 3.1 not 3.2

hw/pci-host/gpex-acpi.c static void acpi_dsdt_add_pci_osc(Aml *dev) · line 123

Quoted 2 lines
hw/pci-host/gpex-acpi.c +2 -0
+     * The UUID in _DSM in this context is
+     * {E5C937D0-3553-4D7A-9117-EA4D19C3434D}

This is just five lines earier, I don't think we need it here.

hw/pci-host/gpex-acpi.c static void acpi_dsdt_add_pci_osc(Aml *dev) · line 125

Quoted 2 lines
hw/pci-host/gpex-acpi.c +2 -0
+     */
+    ifctx1 = aml_if(aml_equal(aml_arg(2), aml_int(5)));

add comment:

	/* Arg2: Function Index: 5 */

hw/pci-host/gpex-acpi.c static void acpi_dsdt_add_pci_osc(Aml *dev) · line 126

Quoted 1 line
hw/pci-host/gpex-acpi.c +1 -0
+    aml_append(ifctx1, aml_return(aml_int(0)));

add comment: /* 0 - do not ignore ... (quote spec I don't have it to hand) */

Quoted 6 lines
hw/pci-host/gpex-acpi.c +1 -0
+    aml_append(ifctx, ifctx1);
     aml_append(method, ifctx);
 
     byte_list[0] = 0;
Signature · 1 line
2.28.0
author
#7 lore raw

Hi Michael,

Quoted 50 lines · Michael S. Tsirkin
Quoted 35 lines · Jiahui Cen

There may be some differences in pci resource assignment between guest os and firmware.

Eg. A Bridge with Bus [d2]

    -+-[0000:d2]---01.0-[d3]----01.0

    where [d2:01.00] is a pcie-pci-bridge with BAR0 (mem, 64-bit, non-pref) [size=256]
          [d3:01.00] is a PCI Device with BAR0 (mem, 64-bit, pref) [size=128K]
                                          BAR4 (mem, 64-bit, pref) [size=64M]

    In EDK2, the Resource Map would be:
        PciBus: Resource Map for Bridge [D2|01|00]
        Type = PMem64; Base = 0x8004000000;     Length = 0x4100000;     Alignment = 0x3FFFFFF
           Base = 0x8004000000; Length = 0x4000000;     Alignment = 0x3FFFFFF;  Owner = PCI [D3|01|00:20]
           Base = 0x8008000000; Length = 0x20000;       Alignment = 0x1FFFF;    Owner = PCI [D3|01|00:10]
        Type =  Mem64; Base = 0x8008100000;     Length = 0x100; Alignment = 0xFFF

    While in Linux, kernel will use 0x2FFFFFF as the alignment to calculate
    the PMem64 size, which would be 0x6000000.

The diffences could result in resource assignment failure.

Using _DSM #5 method to inform guest os not to ignore the PCI configuration that firmware has done at boot time could handle the differences.

---
 hw/pci-host/gpex-acpi.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)
hw/pci-host/gpex-acpi.c +0 -0
index 071aa11b5c..2b490f3379 100644

hw/pci-host/gpex-acpi.c static void acpi_dsdt_add_pci_osc(Aml *dev) · line 116

Quoted 6 lines
hw/pci-host/gpex-acpi.c +1 -1
@@ -112,10 +112,19 @@ static void acpi_dsdt_add_pci_osc(Aml *dev)
     UUID = aml_touuid("E5C937D0-3553-4D7A-9117-EA4D19C3434D");
     ifctx = aml_if(aml_equal(aml_arg(0), UUID));
     ifctx1 = aml_if(aml_equal(aml_arg(2), aml_int(0)));
-    uint8_t byte_list[1] = {1};
+    uint8_t byte_list[1] = {0x21};
     buf = aml_buffer(1, byte_list);

Hmm what is this change for?

I also noticed something weird.
Spec seems to say for _DSM for PCI Express Slot Information:

Quoted 9 lines

Arguments:
Arg0: UUID: E5C937D0-3553-4d7a-9117-EA4D19C3434D
Arg1: Revision ID: 2
Arg2: Function Index: 1
Arg3: Empty Package

how come we are comparing function index to 0 here?

PCI Firmware Spec says in 4.6.1. _DSM for PCI Express Slot Information

Note: Function 0 is a generic Query function that is supported by _DSMs with any UUID and Revision ID. The definition of function 0 is generic to _DSM and specified in the ACPI Specification, Version 3.0 (or later).

And ACPI Spec says in 9.1.1 _DSM (Device Specific Method)

Return Value Information:
If Function Index is zero, the return is a buffer containing one bit for each function index, starting with zero. Bit 0 indicates whether there is support for any functions other than function 0 for the specified UUID and Revision ID. If set to zero, no functions are supported (other than function zero) for the specified UUID and Revision ID. If set to one, at least one additional function is supported. For all other bits in the buffer, a bit is set to zero to indicate if that function index is not supported for the specific UUID and Revision ID. (For example, bit 1 set to 0 indicates that function index 1 is not supported for the specific UUID and Revision ID.)

I have no idea whether the original code does aim to use _DSM #0 by setting function index 0 (The return value seems not to be suitable with _DSM #1). But if it does, I think it is necessary to set bit 5 in return value to indicate _DSM #5 function is supported.

Quoted 4 lines

Also, as long as we are changing this probably shouldn't hard-code 1 as array size ...

Is a macro enough? Like #define RET_BUF_SIZE 2

Quoted 7 lines

hw/pci-host/gpex-acpi.c static void acpi_dsdt_add_pci_osc(Aml *dev) · line 121

Quoted 6 lines
hw/pci-host/gpex-acpi.c +3 -0
     aml_append(ifctx1, aml_return(buf));
     aml_append(ifctx, ifctx1);
+
+    /* PCI Firmware Specification 3.2
+     * 4.6.5. _DSM for Ignoring PCI Boot Configurations
Quoted 6 lines

Note you must always quote the most recent spec that your change refers to. This is so people can figure out legacy guest compatibility.

In this case I think this first appeard in 3.1 not 3.2

OK, I'll fix this.

Quoted 5 lines

hw/pci-host/gpex-acpi.c static void acpi_dsdt_add_pci_osc(Aml *dev) · line 123

Quoted 2 lines
hw/pci-host/gpex-acpi.c +2 -0
+     * The UUID in _DSM in this context is
+     * {E5C937D0-3553-4D7A-9117-EA4D19C3434D}

This is just five lines earier, I don't think we need it here.

Will remove.

Quoted 5 lines

hw/pci-host/gpex-acpi.c static void acpi_dsdt_add_pci_osc(Aml *dev) · line 125

Quoted 2 lines
hw/pci-host/gpex-acpi.c +2 -0
+     */
+    ifctx1 = aml_if(aml_equal(aml_arg(2), aml_int(5)));

add comment:

	/* Arg2: Function Index: 5 */

Will add.

Quoted 6 lines

hw/pci-host/gpex-acpi.c static void acpi_dsdt_add_pci_osc(Aml *dev) · line 126

Quoted 2 lines
hw/pci-host/gpex-acpi.c +1 -0
+    aml_append(ifctx1, aml_return(aml_int(0)));

add comment: /* 0 - do not ignore ... (quote spec I don't have it to hand) */

Will add.

Thanks,
Jiahui

Quoted 12 lines
Quoted 9 lines
hw/pci-host/gpex-acpi.c +1 -0
+    aml_append(ifctx, ifctx1);
     aml_append(method, ifctx);
 
     byte_list[0] = 0;
Signature · 1 line
2.28.0

.

#8 lore raw
Quoted 81 lines · Jiahui Cen

Hi Michael,

Quoted 50 lines · Michael S. Tsirkin
Quoted 35 lines · Jiahui Cen

There may be some differences in pci resource assignment between guest os and firmware.

Eg. A Bridge with Bus [d2]

    -+-[0000:d2]---01.0-[d3]----01.0

    where [d2:01.00] is a pcie-pci-bridge with BAR0 (mem, 64-bit, non-pref) [size=256]
          [d3:01.00] is a PCI Device with BAR0 (mem, 64-bit, pref) [size=128K]
                                          BAR4 (mem, 64-bit, pref) [size=64M]

    In EDK2, the Resource Map would be:
        PciBus: Resource Map for Bridge [D2|01|00]
        Type = PMem64; Base = 0x8004000000;     Length = 0x4100000;     Alignment = 0x3FFFFFF
           Base = 0x8004000000; Length = 0x4000000;     Alignment = 0x3FFFFFF;  Owner = PCI [D3|01|00:20]
           Base = 0x8008000000; Length = 0x20000;       Alignment = 0x1FFFF;    Owner = PCI [D3|01|00:10]
        Type =  Mem64; Base = 0x8008100000;     Length = 0x100; Alignment = 0xFFF

    While in Linux, kernel will use 0x2FFFFFF as the alignment to calculate
    the PMem64 size, which would be 0x6000000.

The diffences could result in resource assignment failure.

Using _DSM #5 method to inform guest os not to ignore the PCI configuration that firmware has done at boot time could handle the differences.

---
 hw/pci-host/gpex-acpi.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)
hw/pci-host/gpex-acpi.c +0 -0
index 071aa11b5c..2b490f3379 100644

hw/pci-host/gpex-acpi.c static void acpi_dsdt_add_pci_osc(Aml *dev) · line 116

Quoted 6 lines
hw/pci-host/gpex-acpi.c +1 -1
@@ -112,10 +112,19 @@ static void acpi_dsdt_add_pci_osc(Aml *dev)
     UUID = aml_touuid("E5C937D0-3553-4D7A-9117-EA4D19C3434D");
     ifctx = aml_if(aml_equal(aml_arg(0), UUID));
     ifctx1 = aml_if(aml_equal(aml_arg(2), aml_int(0)));
-    uint8_t byte_list[1] = {1};
+    uint8_t byte_list[1] = {0x21};
     buf = aml_buffer(1, byte_list);

Hmm what is this change for?

I also noticed something weird.
Spec seems to say for _DSM for PCI Express Slot Information:

Quoted 9 lines

Arguments:
Arg0: UUID: E5C937D0-3553-4d7a-9117-EA4D19C3434D
Arg1: Revision ID: 2
Arg2: Function Index: 1
Arg3: Empty Package

how come we are comparing function index to 0 here?

PCI Firmware Spec says in 4.6.1. _DSM for PCI Express Slot Information

Note: Function 0 is a generic Query function that is supported by _DSMs with any UUID and Revision ID. The definition of function 0 is generic to _DSM and specified in the ACPI Specification, Version 3.0 (or later).

And ACPI Spec says in 9.1.1 _DSM (Device Specific Method)

Return Value Information:
If Function Index is zero, the return is a buffer containing one bit for each function index, starting with zero. Bit 0 indicates whether there is support for any functions other than function 0 for the specified UUID and Revision ID. If set to zero, no functions are supported (other than function zero) for the specified UUID and Revision ID. If set to one, at least one additional function is supported. For all other bits in the buffer, a bit is set to zero to indicate if that function index is not supported for the specific UUID and Revision ID. (For example, bit 1 set to 0 indicates that function index 1 is not supported for the specific UUID and Revision ID.)

Quoted 4 lines

I have no idea whether the original code does aim to use _DSM #0 by setting function index 0 (The return value seems not to be suitable with _DSM #1). But if it does, I think it is necessary to set bit 5 in return value to indicate _DSM #5 function is supported.

So let's make it self documenting:

{
	0x1 << 0 /* support for support for any functions other than function 0 */ |
	0x1 << 5 /* support for function 5 */
}
Quoted 6 lines
Quoted 4 lines

Also, as long as we are changing this probably shouldn't hard-code 1 as array size ...

Is a macro enough? Like #define RET_BUF_SIZE 2

Better to use

uint8_t byte_list[] = { ... };

And then ARRAY_SIZE(byte_list)

Quoted 56 lines
Quoted 7 lines

hw/pci-host/gpex-acpi.c static void acpi_dsdt_add_pci_osc(Aml *dev) · line 121

Quoted 6 lines
hw/pci-host/gpex-acpi.c +3 -0
     aml_append(ifctx1, aml_return(buf));
     aml_append(ifctx, ifctx1);
+
+    /* PCI Firmware Specification 3.2
+     * 4.6.5. _DSM for Ignoring PCI Boot Configurations
Quoted 6 lines

Note you must always quote the most recent spec that your change refers to. This is so people can figure out legacy guest compatibility.

In this case I think this first appeard in 3.1 not 3.2

OK, I'll fix this.

Quoted 5 lines

hw/pci-host/gpex-acpi.c static void acpi_dsdt_add_pci_osc(Aml *dev) · line 123

Quoted 2 lines
hw/pci-host/gpex-acpi.c +2 -0
+     * The UUID in _DSM in this context is
+     * {E5C937D0-3553-4D7A-9117-EA4D19C3434D}

This is just five lines earier, I don't think we need it here.

Will remove.

Quoted 5 lines

hw/pci-host/gpex-acpi.c static void acpi_dsdt_add_pci_osc(Aml *dev) · line 125

Quoted 2 lines
hw/pci-host/gpex-acpi.c +2 -0
+     */
+    ifctx1 = aml_if(aml_equal(aml_arg(2), aml_int(5)));

add comment:

	/* Arg2: Function Index: 5 */

Will add.

Quoted 6 lines

hw/pci-host/gpex-acpi.c static void acpi_dsdt_add_pci_osc(Aml *dev) · line 126

Quoted 2 lines
hw/pci-host/gpex-acpi.c +1 -0
+    aml_append(ifctx1, aml_return(aml_int(0)));

add comment: /* 0 - do not ignore ... (quote spec I don't have it to hand) */

Will add.

Thanks,
Jiahui

Quoted 12 lines
Quoted 9 lines
hw/pci-host/gpex-acpi.c +1 -0
+    aml_append(ifctx, ifctx1);
     aml_append(method, ifctx);
 
     byte_list[0] = 0;
Signature · 1 line
2.28.0

.

author
#9 lore raw

Hi Michael,

Quoted 103 lines · Michael S. Tsirkin
Quoted 81 lines · Jiahui Cen

Hi Michael,

Quoted 50 lines · Michael S. Tsirkin
Quoted 35 lines · Jiahui Cen

There may be some differences in pci resource assignment between guest os and firmware.

Eg. A Bridge with Bus [d2]

    -+-[0000:d2]---01.0-[d3]----01.0

    where [d2:01.00] is a pcie-pci-bridge with BAR0 (mem, 64-bit, non-pref) [size=256]
          [d3:01.00] is a PCI Device with BAR0 (mem, 64-bit, pref) [size=128K]
                                          BAR4 (mem, 64-bit, pref) [size=64M]

    In EDK2, the Resource Map would be:
        PciBus: Resource Map for Bridge [D2|01|00]
        Type = PMem64; Base = 0x8004000000;     Length = 0x4100000;     Alignment = 0x3FFFFFF
           Base = 0x8004000000; Length = 0x4000000;     Alignment = 0x3FFFFFF;  Owner = PCI [D3|01|00:20]
           Base = 0x8008000000; Length = 0x20000;       Alignment = 0x1FFFF;    Owner = PCI [D3|01|00:10]
        Type =  Mem64; Base = 0x8008100000;     Length = 0x100; Alignment = 0xFFF

    While in Linux, kernel will use 0x2FFFFFF as the alignment to calculate
    the PMem64 size, which would be 0x6000000.

The diffences could result in resource assignment failure.

Using _DSM #5 method to inform guest os not to ignore the PCI configuration that firmware has done at boot time could handle the differences.

---
 hw/pci-host/gpex-acpi.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)
hw/pci-host/gpex-acpi.c +0 -0
index 071aa11b5c..2b490f3379 100644

hw/pci-host/gpex-acpi.c static void acpi_dsdt_add_pci_osc(Aml *dev) · line 116

Quoted 6 lines
hw/pci-host/gpex-acpi.c +1 -1
@@ -112,10 +112,19 @@ static void acpi_dsdt_add_pci_osc(Aml *dev)
     UUID = aml_touuid("E5C937D0-3553-4D7A-9117-EA4D19C3434D");
     ifctx = aml_if(aml_equal(aml_arg(0), UUID));
     ifctx1 = aml_if(aml_equal(aml_arg(2), aml_int(0)));
-    uint8_t byte_list[1] = {1};
+    uint8_t byte_list[1] = {0x21};
     buf = aml_buffer(1, byte_list);

Hmm what is this change for?

I also noticed something weird.
Spec seems to say for _DSM for PCI Express Slot Information:

Quoted 9 lines

Arguments:
Arg0: UUID: E5C937D0-3553-4d7a-9117-EA4D19C3434D
Arg1: Revision ID: 2
Arg2: Function Index: 1
Arg3: Empty Package

how come we are comparing function index to 0 here?

PCI Firmware Spec says in 4.6.1. _DSM for PCI Express Slot Information

Note: Function 0 is a generic Query function that is supported by _DSMs with any UUID and Revision ID. The definition of function 0 is generic to _DSM and specified in the ACPI Specification, Version 3.0 (or later).

And ACPI Spec says in 9.1.1 _DSM (Device Specific Method)

Return Value Information:
If Function Index is zero, the return is a buffer containing one bit for each function index, starting with zero. Bit 0 indicates whether there is support for any functions other than function 0 for the specified UUID and Revision ID. If set to zero, no functions are supported (other than function zero) for the specified UUID and Revision ID. If set to one, at least one additional function is supported. For all other bits in the buffer, a bit is set to zero to indicate if that function index is not supported for the specific UUID and Revision ID. (For example, bit 1 set to 0 indicates that function index 1 is not supported for the specific UUID and Revision ID.)

Quoted 4 lines

I have no idea whether the original code does aim to use _DSM #0 by setting function index 0 (The return value seems not to be suitable with _DSM #1). But if it does, I think it is necessary to set bit 5 in return value to indicate _DSM #5 function is supported.

So let's make it self documenting:

{
	0x1 << 0 /* support for support for any functions other than function 0 */ |
	0x1 << 5 /* support for function 5 */
}
Quoted 5 lines
Quoted 4 lines

Also, as long as we are changing this probably shouldn't hard-code 1 as array size ...

Quoted 8 lines
Quoted 1 line

Is a macro enough? Like #define RET_BUF_SIZE 2

Better to use

uint8_t byte_list[] = { ... };

And then ARRAY_SIZE(byte_list)

Got it. I'll fix in the next patch.

Thanks,
Jiahui

Quoted 60 lines
Quoted 37 lines
Quoted 8 lines

hw/pci-host/gpex-acpi.c static void acpi_dsdt_add_pci_osc(Aml *dev) · line 121

Quoted 7 lines
hw/pci-host/gpex-acpi.c +3 -0
     aml_append(ifctx1, aml_return(buf));
     aml_append(ifctx, ifctx1);
+
+    /* PCI Firmware Specification 3.2
+     * 4.6.5. _DSM for Ignoring PCI Boot Configurations
Quoted 6 lines

Note you must always quote the most recent spec that your change refers to. This is so people can figure out legacy guest compatibility.

In this case I think this first appeard in 3.1 not 3.2

OK, I'll fix this.

Quoted 5 lines

hw/pci-host/gpex-acpi.c static void acpi_dsdt_add_pci_osc(Aml *dev) · line 123

Quoted 2 lines
hw/pci-host/gpex-acpi.c +2 -0
+     * The UUID in _DSM in this context is
+     * {E5C937D0-3553-4D7A-9117-EA4D19C3434D}

This is just five lines earier, I don't think we need it here.

Will remove.

Quoted 5 lines

hw/pci-host/gpex-acpi.c static void acpi_dsdt_add_pci_osc(Aml *dev) · line 125

Quoted 2 lines
hw/pci-host/gpex-acpi.c +2 -0
+     */
+    ifctx1 = aml_if(aml_equal(aml_arg(2), aml_int(5)));

add comment:

	/* Arg2: Function Index: 5 */

Will add.

Quoted 4 lines
Quoted 2 lines
hw/pci-host/gpex-acpi.c +1 -0
+    aml_append(ifctx1, aml_return(aml_int(0)));
Quoted 20 lines
Quoted 2 lines

add comment: /* 0 - do not ignore ... (quote spec I don't have it to hand) */

Will add.

Thanks,
Jiahui

Quoted 12 lines
Quoted 9 lines
hw/pci-host/gpex-acpi.c +1 -0
+    aml_append(ifctx, ifctx1);
     aml_append(method, ifctx);
 
     byte_list[0] = 0;
Signature · 1 line
2.28.0

.

.

#10 lore raw
In reply to Jiahui Cen

Hi,

Quoted 20 lines · Jiahui Cen

There may be some differences in pci resource assignment between guest os and firmware.

Eg. A Bridge with Bus [d2]

    -+-[0000:d2]---01.0-[d3]----01.0

    where [d2:01.00] is a pcie-pci-bridge with BAR0 (mem, 64-bit, non-pref) [size=256]
          [d3:01.00] is a PCI Device with BAR0 (mem, 64-bit, pref) [size=128K]
                                          BAR4 (mem, 64-bit, pref) [size=64M]

    In EDK2, the Resource Map would be:
        PciBus: Resource Map for Bridge [D2|01|00]
        Type = PMem64; Base = 0x8004000000;     Length = 0x4100000;     Alignment = 0x3FFFFFF
           Base = 0x8004000000; Length = 0x4000000;     Alignment = 0x3FFFFFF;  Owner = PCI [D3|01|00:20]
           Base = 0x8008000000; Length = 0x20000;       Alignment = 0x1FFFF;    Owner = PCI [D3|01|00:10]
        Type =  Mem64; Base = 0x8008100000;     Length = 0x100; Alignment = 0xFFF

    While in Linux, kernel will use 0x2FFFFFF as the alignment to calculate
    the PMem64 size, which would be 0x6000000.
Quoted 6 lines

The diffences could result in resource assignment failure.

Using _DSM #5 method to inform guest os not to ignore the PCI configuration that firmware has done at boot time could handle the differences.

Since this patch is in qemu (ie starting with qemu v6.0), some of my qemu tests booting aarch64 systems with efi bios no longer work. For example, the following command fails to instantiate the Ethernet interface.

CMDLINE="root=/dev/vda console=ttyAMA0"
ROOTFS="rootfs.ext2"

qemu-system-aarch64 -M virt -kernel arch/arm64/boot/Image -no-reboot \
	-m 512 -cpu cortex-a57 -no-reboot \
	-device tulip,netdev=net0 -netdev user,id=net0 \
	-bios QEMU_EFI-aarch64.fd \
	-snapshot \
	-device virtio-blk-device,drive=d0 \
	-drive file=${ROOTFS},if=none,id=d0,format=raw \
	-nographic -serial stdio -monitor none \
	--append "${CMDLINE}"

QEMU_EFI-aarch64.fd is from https://retrage.github.io/edk2-nightly/.

Key difference is PCI BAR assignment.

good (qemu v5.2):

[    3.921801] pci 0000:00:01.0: [1011:0019] type 00 class 0x020000
[    3.922207] pci 0000:00:01.0: reg 0x10: [io  0x0000-0x007f]
[    3.922505] pci 0000:00:01.0: reg 0x14: [mem 0x10000000-0x1000007f]
[    3.927111] pci 0000:00:01.0: BAR 0: assigned [io  0x1000-0x107f]
[    3.927455] pci 0000:00:01.0: BAR 1: assigned [mem 0x10000000-0x1000007f]

bad (qemu v6.0 and later):

[    3.922887] pci 0000:00:01.0: [1011:0019] type 00 class 0x020000
[    3.923278] pci 0000:00:01.0: reg 0x10: [io  0x0000-0x007f]
[    3.923451] pci 0000:00:01.0: reg 0x14: [mem 0x10000000-0x1000007f]

Reverting this patch fixes the problem.

Does this mean that I can no longer run aarch64 efi boot tests with qemu v6.0 and later if the test involves PCI devices ? Or is there some alternative command line which would still work ?

Thanks,
Guenter